Abstract
Objective
This study sought to characterize how a population experienced health-related social
needs (HRSNs) over time.
Methods
We employed hidden Markov modeling using data extracted from a natural language processing
state machine from 2018 to 2020 to examine whether a patient experienced any food,
legal, transportation, employment, financial, or housing needs. Characteristics of
patients transitioning into low/high-risk states were compared. We also identified
the frequency at which patients transitioned according to their risk state.
Results
Our results identified that five hidden states best represented how patients are experiencing
HRSNs longitudinally. Of 48,055 patients, 80% were categorized in states 1 and 2,
labeled as low risk. Nine percent, 8%, and 3% of the study population were labeled
as medium, high, and very high risk, respectively. Results also showed that low and
high-risk patients (states 1, 2, and 5) only transition states once every year and
a half, while patients in medium and high-risk states transition approximately once
per year.
Conclusion
Low and very high-risk patients tend to remain in the same state over time, suggesting
that low-risk patients may have the means to maintain a healthy state while very high-risk
patients have a difficult time resolving multiple HRSNs. Early screening and immediate
interventions may be beneficial in mitigating the persistent harm of unaddressed HRSNs.
Keywords
social factors - health-related social needs - social determinants of health - hidden
Markov model - natural language processing - electronic health records
